Methods and apparatus to self-guardrail large language model responses

An apparatus, comprising:

• memory;
• machine readable instructions;
• and
• at least one processor circuit to at least one of instantiate or execute the machine readable instructions to:
• access a request to validate whether a digital message is at least one of a spam message, a phishing message, or a malicious message;
• provide, to a first large language model, an initial prompt based on the digital message to generate a response message including an explanation of whether the digital message is at least one of the spam message, the phishing message, or the malicious message;
• perform a first-pass screening of the response message to determine whether the response message includes at least one word or phrase identified as inappropriate;
• when the first-pass screening indicates the response message does not include a word or phrase identified as inappropriate, perform a second determination by causing a second large language model to evaluate the response message with respect to at least one hypothesis to determine whether the response message is inappropriate;
• and
• when the response message is determined to be inappropriate based on the second determination, modify the initial prompt to create a modified prompt by insertion of an anti-hypothesis corresponding to the at least one hypothesis, and provide the modified prompt to the first large language model to trigger generation of a subsequent response message including a subsequent explanation of whether the digital message is at least one of the spam message, the phishing message, or the malicious message.